Block unveils Dessa, a locally run voice model that sounds strikingly human

Block, the fintech company led by Jack Dorsey, has released a new voice model called Dessa from its research lab. The model reproduces human expressions such as laughs, pauses and cries, and can run locally rather than on remote servers. The company announced it on October 23.

How the model works

Block says Dessa was trained entirely on human speech using a single NVIDIA GPU. It is a full-duplex model, meaning it can speak and listen simultaneously, similar to the GPT-Live model OpenAI released earlier this year.

Speech models are typically paired with a large language model that supplies knowledge. Block says Dessa's knowledge comes from the conversational data it was trained on.

The company argues that current voice AI can sound unnatural because text sits in the pipeline and drives speech generation. Models such as ElevenLabs v4 or Fish Audio Drama 3 use voice tags to steer emotion. Block claims emotion is built into Dessa itself, although it is unclear whether tags steer it in some way.

In a blog post, the company said converting speech to text discards a great deal of information and that the most powerful AIs cannot hear how a person feels.

Examples and availability

Block's researchers posted several sample conversations with different voices, cadences and emotions. In one, the bot said it was tired from work and sounded tired. Another shows a Block team member playing a game with the voice assistant.

The model can run on a consumer-grade computer such as a MacBook M4 Max or better. There is no public access yet, and Block has called on safety researchers to test it, likely to examine cases where a voice agent could deceive the people it talks to.

Block operates in finance and sells products to small businesses. Its Square platform already uses voice AI to handle ordering. A proprietary on-device voice model could power more business and consumer services at lower cost.